The Agentic platform allows you to create your own agents, customized to the needs of your teams, with each agent having specific instructions and tools to use.

Overview

An agent is comprised of the following elements:

• A name, used to call the agent in the chat with “@”;

• An avatar (or agent image) that will be displayed next to the agent’s name in the chat;

• A brief description of what the agent is used for. This description will be displayed on the agent’s presentation card and may help users choose an agent;

• A category that designates the context in which the agent is mostly used.

• A series of instructions defining the role of the agent, its tasks, what data it has access to, etc. The purpose of the instructions is to guide the agent in the way they should answer;

• A model (also known as an LLM or Large Language Model), which is used to retrieve relevant information and generate the answer to the user’s query.

• One or more data sources, including documents and external integrations (such as websites or knowledge bases), that the agent should utilize to deliver accurate and relevant information.

• One or several tools that determine what an agent can do;

• One or several toolsets, which are groups of tools and/or data sources that determine what an agent can do and what data it has access to;

• Guardrails, which are predefined guidelines or rules ensuring that an agent operates safely and effectively.

Build your first agent

Determine your agent’s purpose

Before you create and set up your agent, it's important to consider a few key factors. Start by defining your use case and how the agent will be utilized. Here are some essential questions to help you configure your agent effectively:

• What is the primary purpose of my agent?

• What persona should it adopt?

• Who will be using the agent?

• What tasks will it need to perform?

• What data should it access?

Create a new agent

To create an agent starting from the home page, navigate to the Agents menu item, then click the Create an agent button.

You can either create a new agent from scratch or by using a template. In this example, we will create an agent from scratch by clicking the New agent/Start from scratch button.

General information

Begin by providing general information about your agent, such as:

• its name, such as Customer Support for example. The agent’s name must be short and descriptive at the same time.

• a dedicated agent image. To edit the image, click the default logo to open the panel on the left where you can change the background color and choose an icon among the ones provided.

• a category, which will make it easier for users to find agents relevant to their use case. For instance, an agent with the HR type means that it is used for HR tasks.

• a short description of its purpose.

• the instructions your agents should follow such as role, its tasks, what data it has access to, etc. You can write the instructions yourself or with the help of AI by clicking the Generate with AI button above the text area.

• the AI model to use for your agent. Choose one amongst the ones proposed in the list.

• the guardrails to apply to your agent. You can choose several from the ones proposed.

Data sources

Next, you should select the data source(s) your agent should have access. Depending on the agent’s purpose, adding data sources might not be needed.

You have the option of adding a document by clicking the Add a document button or linking an external integration by clicking the Link an external integration button.

Tools

You can add one or more tools to your agent by clicking the Add a tool button.

Define the visibility of your agent

Finally, it’s time to decide who can access your agent.

You can share your agent with specific people and teams or give everyone access to use or edit the agent.

• Private: the agent is only available to you.

• Company: the agent is available to you and your colleagues company-wide.

• Shared: the agent is available to specific users designated by you.

– example screenshot to do --

Test your agent

Once you finished the configuration of your agent, you can test it by typing in the chat box in the right panel.